Frans Willekens discusses multi-staged decision processes over the human life course. His post is based on the Population Studies special issue dedicated to the subject published in October 2017.
1.The importance of individual agency for demography
Population change is determined by the life choices individuals and families make. To understand population change, we need to understand how choices are made. Critical decisions in life, such as the decision to marry, to have a child to migrate or to retire, are outcomes of cognitive and emotional processes. These processes are complex, in particular when individuals have the freedom of choice, making outcomes difficult to predict. Furthermore, these processes take time, which is often in short supply when collecting and interpreting information, evaluating the options, and reaching a decision. Time constraints (deadlines) may be self-imposed or imposed biologically (e.g., menopause), socially (e.g., age norms), or legally (e.g. stage of pregnancy at abortion). Time, and other resources and constraints differ between persons. The differences are amplified when some persons are risk-averse and others risk-seeking, and mechanisms of risk sharing and social support vary greatly. Because of all these differences, people’s lives and biographies are diverse, and population change is colourful but complex. To deal with these complexities, formal demography should shift its focus from demographic events to the processes that generate the events. Social demography should move beyond statistical associations (regression models) and concentrate on the causal mechanisms. A positive consequence of such changes will be a blurring of the distinction between formal and social demography, and the emergence of a truly comprehensive and coherent population science.
The purpose of this blog post is to situate the subject of decision-making processes over the life course in the broader context of population studies. How would our discipline change if demographers embrace the science of choice? To answer that question, I start from the most basic observation of demography; namely, that population change is a consequence of demographic events. Events are experienced by individual members of the population and they result in attrition or exit, and accession or entry . That fundamental mechanism is described by the balancing equation, discussed in Section 2, which links population change to the events causing the change. Although the balancing equation is a model of a population, risk levels are properties of individuals. Risk levels vary by exposure status and they are influenced by personal attributes and contextual factors. Individuals are not passive recipients of risk levels, allocated on the basis of exposure status, personal attributes and contextual factors. Individuals have agency, i.e. the capacity to choose and act accordingly, even if that capacity is often constrained. Demographic models do not account for individual agency, but they should. Micro-level processes of decision making and social interaction should be incorporated into demographic models. That can be done by replacing rates with rules governing the processes leading to demographic events. Micro-level processes of decision making are the subject of Sections 3 and 4. Section 5 discusses briefly how these processes can be modelled. Section 6 concludes this blog post.
2. Linking the individual and population levels through the balancing equation
The balancing equation is an accounting equation that is a basic ingredient of the life table and population projections. In the life table, the population consists of a single cohort and is stratified by age, or by age and personal attributes. Age and personal attributes are individual characteristics, but they also define subpopulations. In the life table, the balancing equation relates changes in cohort size and composition to attritions (decrements) and transitions between subpopulations. Transitions are associated with changes in personal attributes. In the cohort-component projection model, one of the most common demographic forecasting techniques, the population consists of multiple cohorts stratified by age and personal attributes. In projections, the balancing equation is used to relate changes in cohort size and composition to attritions (death) and transitions, and generates new cohorts through accessions (births). Births, deaths and transitions are events experienced by individuals but with consequences at the population level. Two important aspects of population change can be emphasized from the balancing equation: the primacy of the individual and differences between individuals (heterogeneity). The first derives from the fact that the events that cause population change occur at the individual level – and should therefore be studied at that level! The primacy of the individual is accounted for by relating population structure to personal attributes and approaching the events as personal experiences. The second aspect, heterogeneity, can be accounted for by stratification and regression modelling.
Formal demography is built on two key concepts: risk and exposure. Social demography concentrates on the documentation, explanation, and prediction of risk levels. Risk levels vary with age, personal attributes and contextual factors. Contextual factors include institutional factors, such as the household type, political regime, cultural and religious institutions, type of welfare state and health system. Statistical methods, in particular regression models, are used to determine the co-variation of risk levels (hazard rates and probabilities) and mentioned factors. This mode of explanation, often referred to as explanation by statistical association, is currently dominant in demography.
Formal demography disregards population heterogeneity, except for differences in age and sex. Additional differences can be introduced by hazard models with covariates (and other sources of heterogeneity) (see e.g. Mooyaart and Liefbroer, 2016 for a discrete-time hazard model and Grace and Sweeney, 2014 for a continuous-time hazard model used to predict first marriage rates and the expected age at first marriage). By adding unobserved heterogeneity and individual stochasticity, the model predicts event rates and ages at events for individuals. Individual risk levels and individual-based models can be used to track individuals in a population. Microsimulation is a well-known application of individual-based models.
3. Agency and agents (actors)
Individuals considered in Section 2 lack an important property: agency. In demographic models, transitions in the life course are determined by exposure status and risk levels (or rates), often dependent on personal characteristics and contextual factors. However, they are unrelated to biological and cognitive processes internal to the individuals, and unrelated to the social processes invoked by interactions between individuals and between individuals and institutions. That is a major shortcoming.
Agency is the capacity of individuals and institutions to make choices freely and act accordingly and independently. The term actor or agent is usually used to denote an entity with agency. The effect of personal characteristics and contextual factors on behaviour is mediated by an actor’s agency. If an actor lacks the capacity to make choices and act accordingly, the response to mentioned factors will be different from those of actors with unrestricted agency. Agency requires resources, not only cognitive capabilities. Empowerment gives people the capability to act independently and make choices. Agency also depends on the social context.
In 1957 Herbert Simon proposed the concepts of bounded rationality and satisficing behaviour (a combination of satisfy and suffice). These concepts reflect that people do not have the cognitive capabilities and the resources required to choose the option that maximizes benefits. People cannot access all the information needed, and if they could, their minds would be unable to process it properly. The human mind, as Simon put it, is bounded by “cognitive limits”. In Simon’s view, an individual decision is rational if an individual deliberates properly, given the limitations and constraints. That concept of rationality, which has been referred to as procedural rationality, differs from the concept used in neoclassical economics. Satisficing behaviour may be sub-optimal in a neo-classical utility maximization framework, but it is more realistic because it does not rely on the unrealistic assumptions that people maximize the utility of their actions with perfect foresight and stable preferences, and that resources are perfectly mobile and information flows freely and is available to everyone without restriction. The rational choice theory of neoclassical economics may apply to the very few with the agency and environment postulated by the theory, but it does not apply to most people. Nevertheless, rational choice theory flourished in economics. It became popular in sociology too, after James Coleman adopted the economists’ theory of rational choice in his Foundations of social theory. Coleman admits that he adopted that theory because of its simplicity and not because it can be considered realistic. Rational choice theory has been the subject of intense debates. The theory evolved into behavioural economics, which acknowledges the limitations of people and their environment and pictures more realistic agents. Discrete choice models, which were initially based on rational choice theory, evolved into hybrid choice models, which incorporate agency and insights from psychology and sociology. Agency is also at the heart of the capability approach, advocated by Sen, Nussbaum and others. It states that the freedom of choice is to be understood in terms of people’s capabilities, that is, their real opportunities to do and be what they have reason to value.
Agency can be incorporated in the explanation of demographic behaviour by considering how actors make decisions and how decisions lead to action. Decision-making is a process that usually takes time. Decision making can be fast, when mental shortcuts, rules of thumb, or heuristics are used, but major decisions in life are usually slow. The decision to cohabit or to marry, to separate, to change a lifestyle or to migrate, and end-of-life decisions take time. Actions do not follow decisions automatically. During the decision process, others may intervene or conditions may change and influence the decision process and the implementation of a decision.
4. Agents: actions and interactions
The incorporation of agency in demographic explanation calls for a shift from regression models (and explanation by statistical association) to process models (and mechanism-based explanation). A mechanism is a process that connects cause and effect. Social, economic and other factors do not influence behaviour directly, but indirectly, through their influence on actors, the decisions they make and the actions they take. Mechanism-based explanation is actor-based. Some decision processes are relatively simple, e.g. routine behaviour, while others are complex, e.g. life choices. The decision processes that lead to life choices are developmental processes. A developmental process consists of stages and each stage builds on previous stages. Once initiated, a decision process evolves until it results in a choice. A decision process is initiated by changes in conditions, personal attributes, or external events. They may also be motivated by the accumulation of dissatisfaction with the current situation, as may be the case for divorce or outmigration. The first stage of any decision process is to determine what the options (alternatives) are. That involves the collection of information, which depends on cognitive capabilities and social capital, but also on financial resources. People may involve brokers, e.g. marriage brokers in cultures where arranged marriages are customary, and migration brokers in case of international migration. Actors with adequate resources can make better decisions than actors that lack resources. Hence individual agency is context-sensitive; it depends on many contextual factors, including culture. Agency may also be shared with collaborative deliberation, shared decision-making and joint activities as possible outcomes. Shared agency enables actors to act together effectively. It does not mean, however, that they have equal access to resources or participate equally in the decision-making and the subsequent action.
The modelling of decision processes incorporates the central concepts of exposure, risks and transitions. Decision processes are multi-stage processes and involve transitions between stages. The population at risk at a given age or time is the collection of individuals in the same stage of decision making at that age or time. The time spent in a stage is a random variable, which can be modelled by a waiting time distribution. Approaching decision processes as staging processes in formal demography preserves the centrality of event history methods and opens perspectives for demographers to contribute to decision theories and theories of action. For instance, in their seminal paper “The ‘horse race’ random utility model for choice probabilities and reaction times, and its competing risks interpretation” Marley and Colonius advocated the theory of competing risks to gain better insights in decision processes and their outcomes. Regrettably, that paper did not receive much attention in the behavioural and social sciences.
It is time demographers embrace mechanism-based explanations. The process perspective is familiar to most demographers who study transitions, either at the macro level (fertility and epidemiological transition) or the micro level (life-course transitions). Demographers are particularly well positioned to adopt a process perspective on life choices and the actions that follow. They are well positioned because, in demography, the life course is the dominant paradigm, longitudinal data are relatively common, and event history analysis and sequence analysis are well established. What may still be missing is a solid knowledge of decision theories (beyond rational choice theory) and theories of action.
Unfortunately, decision processes operating at the individual level or, in case of shared decision making, at the group level, are not sufficient to explain population change. Demographic change, such as the first or second demographic transition, is conditional on many actors making similar decisions and taking similar actions. Actors may make similar decisions independently, but that is unlikely. More likely is that actors interact and exchange values, information, knowledge and/or resources that motivate other actors to decide and act similarly. That transmission process is a diffusion process. Social learning and social influence are outcomes of a diffusion mechanism, as proposed by Montgomery and Casterline, and later elaborated in the book Diffusion processes and fertility transition. Social learning is the process through which individuals gain knowledge from others, while social influence is the process through which some individuals exert control over others. Diffusion is more rapid if actors communicate frequently, which they are more likely to do if they are neighbours or members of a social network, or if technology or other interventions remove the barriers to communication. Effective communication increases the pace of the diffusion process. The interpretation of cohort effects may be enriched by adding that communication and diffusion during formative years are as important as the similarity of experiences.
5. Modelling decision and diffusion processes
Decision and diffusion processes are micro-level processes, but they generate the patterns and dynamics observed at the macro level. The actors in these processes are individuals and institutions, with some institutions positioned at a micro level (e.g. households) and other at a high level of aggregation, but with actions at different levels (e.g. international organizations with activities at the global, national and local level). This complexity calls for models that can capture decision and diffusion processes realistically. How do such models differ from conventional demographic models? In conventional demographic models, event counts and expected durations (e.g. life expectancy) are predicted from populations at risk (exposure) and risk levels (or event rates). These models are rate-based. Furthermore, the life table and projection models are also rate-based models. In order to incorporate decision and diffusion processes in demographic models, rates should be replaced by rules: decision rules and rules governing social transaction and social diffusion.
Example 1: Job and partner search
Consider an individual in need of a job who is willing to migrate in order to obtain a job. He or she decides to first look for a job locally, but to migrate if no satisfactory job offer arrives within one year. A job has a set of characteristics. Consider a single characteristic, the wage, and postulate a simple decision rule: a job offer is accepted if the wage exceeds a given figure. From the distribution of job offers during a year and their associated wages, one can determine the probability of a job offer that comes with an acceptable wage. Assume that the number of job seekers who find the wage acceptable exceeds the number of vacancies and that a simple rule is applied to distribute job offers among job seekers. The distribution can be on a first come first served basis, at random, or a more complex rule can be postulated. Assume that jobs are allocated randomly to job seekers who find the wage acceptable. Our job seeker migrates if during a 12 month period all job offers go to other persons. In this model, migration is not determined by the number of people at risk (job seekers) and the risk levels (migration rates), but by the number of people at risk, the number of job offers within a year and the associated wages, and the decision rule used by job seekers and the decision rule used by employers. The model is a rule-based model. The numbers of migrations predicted by the rule-based model should be comparable to those predicted by a rate-based model. The decision rule used in this section is derived from the job search theory of labour economics. The minimum acceptable wage is known as the reservation wage.
Suppose that, during the year, our job seeker observes that peers in the social network are accepting job offers while he/she does not get a job offer with an acceptable wage, and finds out that those accepting the offers are less picky. Our job seeker learns to lower the aspiration level (reservation wage). It is also likely that the job seeker reduces the aspiration level with the duration of the job search and if peers exercise social pressure to be less picky. This search mechanism can be modelled using decision rules that are adapted as a result of learning and social influence. It is evident that social interaction and the associated exchange of information increases the likelihood that our job seeker accepts a job offer and does not migrate. The job search mechanism with social learning and influence described here has been applied to partner search (Todd and Billari, 2003; Billari et al., 2007; Willekens, 1988).
Example 2: Family planning
Consider another example. The Programme of Action prepared at the International Population and Development Conference organized by the United Nations in Cairo in 1994, states that couples and individuals should have the right to decide freely and responsibly the number, spacing and timing of their children and should have the information and means to do so. In other words, couples and individuals should be capable to exercise a basic human right. The Demographic and Health Surveys collect information on couples and individuals who want to stop or space childbearing but lack the information and/or means to prevent pregnancy (unmet need). Family planning models have been developed to project the requirements needed to reduce the unmet need or achieve desired fertility (e.g. Spectrum). These models are rate-based, with rates estimated from data. I am not aware of an equivalent agent-based model, in which the decision process is modelled. Modelling the decision process could result in a family planning model that includes the accumulation of information (by source of information), the affordability of contraceptives, the access to safe abortion, and the interdependence between the fertility and other domains of life, such as children’s health and education. Anthropological research and survey research may uncover the rules that govern fertility intentions and the actions following the intentions.
An advantage of rule-based models is that the rules facilitate the incorporation of behavioural and social theories in demographic models. In the job search and migration example, a simple rule determined the decision and action, but in real situations rules can be complex and pertain to different domains of life. For instance, during the job search process, our job seeker could meet a marriage partner or experience another intervening event, which makes him/her change initial decision rules. Rule-based models offer a great opportunity to bridge the gap between formal and social demography. A main feature of rule-based models or agent-based or actor-based models (ABM) as they are more widely known, is their focus on actors, actions and interactions (Billari and Prskawetz, 2003; Billari et al., 2006; Grow and van Bavel, 2017; Willekens et al., 2017).
Agent-based models vary greatly in the use of behavioural and social theory (see Klabunde and Willekens, 2016, for variations in agent-based models of migration). Models may combine rates and rules. For instance, a model that approaches migration from a life-course perspective and accounts for the mutual dependence of migration and events in other domains of life must predict transitions in different domains of life simultaneously. Klabunde et al. (2017) present an agent-based model in which some transitions are modelled as outcomes of exposures and risk levels (rates), while other transitions (in this case migration) are modelled as outcome of decision processes. The decision process considered is rooted in the theory of planned behaviour, a widely used theory of action from social psychology and the model is tested empirically.
The Population Studies special issue is an outcome of the workshop “The science of choice”, organized by the IUSSP Panel on Microsimulation and Agent-Based Models and the Max Planck Institute for Demographic Research. The special issue was edited by Jakub Bijak, Anna Klabunde, Alexia Prskawetz and Frans Willekens. The entire publication is open access, a first in the 70-year history of the oldest English language journal concerned exclusively with demography.
Frans Willekens is emeritus professor of Demography at the University of Groningen (the Netherlands), former director of the Netherlands Interdisciplinary Demographic Institute, former research coordinator and research group leader at the Max Planck Institute for Demographic Research, and resting member of the Royal Netherlands Academy of Arts and Sciences. Frans has retired, but retains a strong interest in demography, specifically agent-based modelling and the science of choice.
 In this text, I do not distinguish between institution and organization.
 Simon, H.A (1957) Models of man: social and rational. Wiley, New York.
 In his paper “In defense of unrealistic assumptions.” (Sociological Theory, 16(2), 1998), Kanazawa defends the use of unrealistic assumptions in theory development. In defense of Milton Friedman (economics) and James Coleman (sociology), he argues that a theory’s assumptions always are and ought to be unrealistic (meaning incomplete, which does not mean it is untrue, i.e. false or highly improbable). He agrees that the assumptions on which rational choice theory is based are incomplete “albeit true in a very limited sense” (p. 200).